Classification of Ultrasonography Images of Human Fatty and Normal Livers using GLCM Textural Features

A computer aided diagnosis methodology for identifying pathology of human liver using 11 statistical textural features extracted from ultrasonography of 14 fatty and 28 normal livers is presented. It was found that supervised classification could differentiate these two classes of data while unsupervised learning failed to achieve that. For supervised learning input sets were constructed in two alternative methods –i) Training set containing two third and test set one third of the data (conventional), ii) A representative training set generated using Self Organizing Map where entire data set was treated as test set. Both the inputs were used with and without Principal Component Analysis. The analysis shows that the Multi Layer Perceptron with conventional data set without pre-processing yields better results as compared to other paradigms. Keyword — Fatty liver, Multi-Layer Perceptron, Principal Component Analysis, Self-Organizing Map,Ultrasonography.